.svg)
I was recently invited to Money20/20 Europe to share my experiences of collaborating with the regulator, perhaps demonstrating the despite some best-practice examples existing, the majority of fintechs and financial institutions have not yet had the opportunity to collaborate with their regulator.
And while RegTech pilots and projects have emerged under some leading-light regulators, the majority of financial institutions have historically viewed regulators as bureaucratic control centres rather than enables of innovation.
For years, one of the most persistent myths in technology has been that regulation stifles innovation. It is a convenient narrative, often repeated by those eager to move quickly, unencumbered by the “tedious” work of ensuring systems are safe, transparent, and fit for purpose. But in financial services, where the consequences of failure can impact livelihoods, institutions, and even civil liberties, the mantra of “move fast and break things” is not just inappropriate, it is dangerous.
What is emerging instead is a new model: one where regulators and industry don’t operate in opposition, but in collaboration. And in the UK, the Financial Conduct Authority (FCA) is demonstrating what that future could look like. Our Napier AI / AML Index found that the UK scored one of the best AI / AML Regulation scores of all forty markets ranked, evidencing the positive view of the FCA as an enabler of artificial intelligence (AI) innovation for anti-money laundering (AML).
From rule-maker to co-innovator
The FCA’s innovation initiatives, particularly its sandbox environments represent a shift in the role of the regulator. Rather than acting solely as an enforcer of rules after the fact, the FCA is enabling innovation to be developed within a regulatory framework from the outset.
This is not a free-for-all. Participation is competitive, structured, and purposeful. Firms must register on the FCA’s innovation platform, apply to specific project cohorts, and demonstrate a genuinely novel use case. In the FCA Supercharged Sandbox, for example, over 200 applications were received for just a few dozen places.
Successful applicants gain access to curated datasets, APIs, and crucially, scalable compute infrastructure. Resources that are often difficult to assemble in isolation. But perhaps most valuable is the oversight: each project is guided by both an FCA representative and an industry mentor, ensuring innovation progresses in line with regulatory expectations.
This is collaboration with intent. Not deregulation, but outcomes-based enablement.
Regulator role in network-based detection
AI innovation in financial crime compliance is accelerating rapidly, but it is constrained by two persistent challenges: data fragmentation and computational limits.
At Napier AI, these challenges were front of mind when we entered the FCA’s Supercharged Sandbox. While we have a mature suite of AI models supporting transaction monitoring and screening, we saw an opportunity to push further into network-based detection.
Financial crime rarely occurs in isolation. It unfolds across networks: accounts, institutions, geographies. But detecting these networks within a single bank’s dataset is inherently limited. Patterns are incomplete. Signals are diluted. And the computational cost of tracing complex transaction paths grows exponentially with each additional node.
This creates two fundamental problems:
- Partial visibility: Key elements of a financial crime network may lie outside a single institution’s data.
- Compute constraints: Tracking multi-hop transaction paths at scale requires significant processing power, often forcing compromises in analysis depth.
The FCA Supercharged Sandbox enabled us to address both challenges.
The fluid dynamics of financial crime
With access to shared datasets, parallel compute infrastructure, and regulatory guidance, we were able to test ideas that would have been impractical within normal R&D constraints.
One such approach was targeted pattern mining. Traditional methods rely heavily on broad subgraph analysis and post-processing to eliminate false positives, which is an expensive and time-consuming process. Instead, we applied information theory and domain knowledge to focus directly on high-risk transaction patterns, reducing noise earlier in the process.
Crucially, the sandbox’s parallel compute environment allowed us to operationalise this approach, using frontier expansion techniques to explore transaction networks at scale. This is now being integrated into the Napier AI platform.
Another innovation drew inspiration from an unlikely source: river pollution monitoring.
In environmental science, contamination is often detected downstream, even when the source is unknown. We asked: could financial crime be modelled in the same way? Instead of tracing entire networks, could we identify “pollution” signals as disruptions in otherwise typical transaction flows?
The answer was yes.
By treating financial systems as dynamic flows, we were able to detect anomalous patterns—such as sudden changes in account behaviour or unusual fund movements—that indicated the presence of illicit activity moving through the network. Even with partial visibility, we could identify fragments of criminal chains as they passed through the dataset.
Importantly, this approach was developed alongside explainability requirements, ensuring outputs could be understood and actioned by analysts, and ultimately incorporated into suspicious activity reporting.
Lessons from the FCA Supercharged Sandbox
Innovation under regulatory collaboration is not without its challenges.
- Prioritise People: It is resource-intensive. Even with data and infrastructure provided, delivering meaningful outcomes within a fixed cohort timeline requires sustained focus and dedicated expertise.
- Be Prepared: While the sandbox supports experimentation, entering with a well-defined problem and hypothesis significantly increases the likelihood of success.
- Keep Trying: Remember, innovation is iterative. Some of our most valuable insights came late in the process, requiring rapid adaptation and refinement of initial ideas.
These are not drawbacks, they are the realities of meaningful innovation, made more effective through structured support.
Regulation as an enabler
It is important to be clear: initiatives like the FCA Supercharged Sandbox do not replace regulatory scrutiny. They do not certify AI models or remove the need for firms to validate their own systems.
What they do instead is shift the focus to outcomes.
The FCA is not dictating how AI should be built, but it is enabling firms to explore what good looks like, and asking the right questions along the way. Are the outputs explainable? Do they align with the institution’s risk profile? Can they be audited?
Accelerating AI for AML
The real significance of the FCA’s innovation platform is not just what it enables today, but what it signals for the future.
Public-private collaboration, when done correctly, does not slow innovation, it accelerates it. It reduces risk not by constraining creativity, but by shaping it. And it ensures that as AI becomes more powerful, it also becomes more accountable.
In financial crime compliance, where the stakes are high and the systems complex, this model is not just beneficial it is essential.
The question is no longer whether regulators and industry can work together.
It is: how quickly can others follow?
